UAG
Universal AI Gateway
Your AI models. Your tools. Your environment.
Quick start · Interfaces · Capabilities · Docs · 日本語 · Other languages
UAG is an open-source, local-first AI agent runtime that connects the language model you choose to the tools and systems you use. Work with code, files, browsers, services, and devices from a consistent interface—without tying your workflows to a single model provider.
| Bring your model | Connect real tools | Stay in control |
|---|---|---|
| Hosted and local models, including OpenAI, Claude, Gemini, Grok, Bedrock, and Ollama. | Files, code, web, MCP, Agent Skills, plugins, cloud services, and IoT. | Local execution, human confirmation, credential storage, and configurable policies. |
UAG is the agent execution layer, not a model provider. External model requests and connected-service calls depend on your configuration; local-first does not mean entirely offline or isolated by default.
Quick start
Requires Python 3.11+. Install from PyPI and launch the interactive CLI:
python -m pip install --upgrade uag
uag
On first launch, the setup wizard helps you choose a provider and configure its credentials. You can also configure providers through environment variables or the credential store.
For a larger set of provider and tool dependencies up front:
python -m pip install "uag[core,providers,tools]"
Then try asking:
> What changed in this Git repository? Summarize the risks.
> Find the latest documentation for this library and explain the migration steps.
> Inspect these files and produce a concise report.
> :help
New to UAG? See the Windows installation walkthrough (including Git and virtual-environment prerequisites) or the configuration reference. Platform-specific integrations are optional.
Interfaces
Use the same agent runtime and tool registry from multiple entry points.
| Interface | Start with | Ideal for |
|---|---|---|
| CLI | uag |
Terminal-first work and automation |
| Desktop | uagg |
Native desktop conversations |
| Web | uagw |
Browser-based conversations and rooms |
| A2A | uaga |
Agent-to-agent integration |
| VS Code | Extension guide | Working alongside code |
Desktop and Web interfaces use additional dependencies that are not included
in the quick-start uag[core,providers,tools] installation:
# Desktop GUI (PySide6 and platform integrations)
python -m pip install "uag[platform]"
# Web UI (web authentication dependencies, FastAPI, and Uvicorn)
python -m pip install "uag[web]" fastapi uvicorn
The Web server can also attempt to install FastAPI and Uvicorn at launch; installing them up front does not depend on runtime auto-installation. See interface options and environment configuration for further setup.
What you can build
| Capability | What UAG provides |
|---|---|
| Model-agnostic workflows | Switch among hosted and local providers using UAGENT_PROVIDER without rewriting tools. |
| Code and file operations | Repository navigation, Git inspection, linting, tests, file operations, and browser automation. |
| Extensible tooling | A catalog of 200+ tools, dynamic tool_catalog / tool_load discovery, custom Python and Rust-backed tools. |
| Connected agents | MCP servers, A2A, Claude Code-compatible plugins, Agent Skills, and sub-agents. |
| Long-running work | Auto-pilot, parallel independent read operations, scheduled tasks, restart recovery, and persisted sessions. |
| Beyond the terminal | Cloud and communication integrations, documents, media, realtime voice, and supported IoT protocols. |
One runtime, many integrations
flowchart LR
UI["CLI · Desktop · Web · VS Code · A2A"] --> UAG["UAG agent runtime"]
UAG --> MODELS["Hosted & local models"]
UAG --> TOOLS["Tools & extensions"]
UAG --> STATE["Sessions & context"]
TOOLS --> SYSTEMS["Files · Code · Web · MCP · Cloud · IoT"]
POLICY["Confirmation · Credentials · Policy"] -.-> UAG
Provider examples: OpenAI, Anthropic, Gemini, Vertex AI, Azure OpenAI, Amazon Bedrock, Grok, DeepSeek, OpenRouter, Ollama, llama.cpp, and LM Studio. For provider-specific settings, see Environment variables.
Customize it with skills and plugins
Extend UAG without modifying the core. Install Agent Skills from SkillsMP or ClawHub, connect MCP servers, or load Claude Code-compatible plugins with tools, skills, agents, and hooks.
:skills mp_search browser automation
:plugin list
:plugin install <source>
:plugin marketplace list
Custom tools can expose TOOL_SPEC and run_tool(); Rust-backed tools are also
supported. Start with the Tool Creator Guide,
Plugin Development Guide, and
MCP OAuth / Proxy Guide.
Built for real workflows
Context and memory
Large tool results can be saved as Artifacts instead of crowding the active
model context. An agent can retrieve the relevant portion later using
artifact_read. Sessions are stored in SQLite by default, with history search
and explicit memory review workflows.
Automation and collaboration
UAG supports parallel independent read operations, multi-agent coordination,
session recovery, and Auto-pilot for multi-round work. Use :auto to start
Auto-pilot; press F12 to interrupt the current response and request that
Auto-pilot stop.
Security and observability
Sensitive operations can require human_ask confirmation. Credentials can be
stored in the operating system's credential manager or an encrypted file
backend. The Web UI supports configured OIDC sign-in and access controls.
OpenTelemetry tracing and metrics are opt-in.
These are configurable controls, not a blanket guarantee that arbitrary tool execution is sandboxed. See the deployment and policy guides before exposing UAG to untrusted users or networks.
Languages, devices, and voice
UAG includes resources for 38 locales (English plus 37 translations), optional full-duplex voice integrations, and tools for systems such as SwitchBot, Matter, ECHONET Lite, BACnet, Modbus TCP, and OPC UA.
Documentation
Browse the documentation directory or read the Japanese README and other translations.
Development and contributing
git clone https://github.com/awaku7/agentcli.git
cd agentcli
python -m pip install -e ".[core,providers,test]"
Before a Python code pull request, run the relevant tests and project checks:
python -m ruff check src tests
python -m black --check src tests
python -m pytest -q .
Read AGENTS.md and the developer guide for detailed contribution guidelines. Issues, documentation improvements, tools, translations, and focused pull requests are welcome.
License
UAG is released under the Apache License 2.0.
1 maintainer
Metadata
1 maintainer
Release files for uag 0.7.33
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | |
|---|---|---|---|---|
| uag-0.7.33.tar.gz | 10.1 MB | Without provenance | Details |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Reset |
|---|---|---|---|---|
| uag-0.7.33-py3-none-any.whl | Python 3 | none | any | Details |
Total release size: 20.8 MB
Release files / uag-0.7.33.tar.gz
| Download URL | uag-0.7.33.tar.gz |
|---|---|
| Size | 10.1 MB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
127e239d5397b933f6e35b74054f5f84b45beea43ee2c2da05fbffccff6a846c
|
|
BLAKE2b-256 checksum How to use checksums |
ed19c41de975701257eb170c2e57d80a83690b2a79e73cbc90976641cf7b7b69
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/7.0.0 CPython/3.14.8
|
Release files / uag-0.7.33-py3-none-any.whl
| Download URL | uag-0.7.33-py3-none-any.whl |
|---|---|
| Size | 10.7 MB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
874abe1eb4918630e8f5f48f5179d36ff4ae98e5ee2ac1fabaa0115acb11336e
|
|
BLAKE2b-256 checksum How to use checksums |
50e340c211212a8155ba1af7c9501ca8d4e7625ac387344fcf539350591f0f4a
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/7.0.0 CPython/3.14.8
|
Security for uag 0.7.33
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
No provenance found
None of the files in this release were uploaded with provenance metadata.
A note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.
